Elsevier

Forensic Science International

Volume 258, January 2016, Pages 81-87
Forensic Science International

Probabilistic age classification with Bayesian networks: A study on the ossification status of the medial clavicular epiphysis

https://doi.org/10.1016/j.forsciint.2015.11.010Get rights and content

Highlights

  • Bayesian methods are suited to deal with uncertainty related to forensic age estimation.

  • They allow users to provide results in form which fulfill the forensic and legal needs.

  • Bayesian networks are powerful in forensic age estimation context.

  • This kind of probabilistic tools ease the application of the Bayesian approach.

  • Exploratory analyses on the collar bones development are promising.

Abstract

In the past few decades, the rise of criminal, civil and asylum cases involving young people lacking valid identification documents has generated an increase in the demand of age estimation. The chronological age or the probability that an individual is older or younger than a given age threshold are generally estimated by means of some statistical methods based on observations performed on specific physical attributes. Among these statistical methods, those developed in the Bayesian framework allow users to provide coherent and transparent assignments which fulfill forensic and medico-legal purposes. The application of the Bayesian approach is facilitated by using probabilistic graphical tools, such as Bayesian networks. The aim of this work is to test the performances of the Bayesian network for age estimation recently presented in scientific literature in classifying individuals as older or younger than 18 years of age. For these exploratory analyses, a sample related to the ossification status of the medial clavicular epiphysis available in scientific literature was used. Results obtained in the classification are promising: in the criminal context, the Bayesian network achieved, on the average, a rate of correct classifications of approximatively 97%, whilst in the civil context, the rate is, on the average, close to the 88%. These results encourage the continuation of the development and the testing of the method in order to support its practical application in casework.

Introduction

Since knowing the age of an individual is necessary for a large number of social and juridical situations (both civil and criminal), the practice of age estimation of living persons has extremely gained in importance in the past few decades [1], [2], [3], [4]. This is mostly due to the increase in cross-border migration movements or human trafficking involving people unable to provide valid documentary evidence for attesting their age [5]. Generally, the range of ages having a particular legal interest is between 14 and 22 years of age, thus it is essential to be able to estimate the chronological age across this interval and/or to decide whether an individual belongs to a given age class included in this range [3], [4]. According to some approaches, such as those proposed by the Study Group on Forensic Age Diagnostic (AGFAD) and the Forensic Anthropology Society of Europe (FASE) [6], age estimation should be appropriately based on the examination of some physical attributes, generally related to bones and dental maturation [4], [7], [8]. The age estimation is usually performed in two steps: first, the degree of maturity reached by a physical indicator is assessed in ordered developmental stages or through the assignment of score values [9], [10], [11]. Then, the observed degree of maturity is converted into the estimated chronological age or in the probability that the individual is included in a meaningful range of ages by means of different statistical methods, usually related to a specific physical attribute. Results should be provided in an appropriate form in order to fulfill the specific forensic and legal needs. From this point of view, the Bayesian approach is particularly suited because it allows the user to coherently deal with the uncertainty related to age estimation and to provide results in a transparent and logical form [12]. Some Bayesian methods allow one to update an initial (or prior) belief about the chronological age of the examined individual in the light of the observed degree of maturity of a given physical attribute [12], [13], [14]. Results are then provided in the form of a posterior probability distribution, which encapsulates all uncertainty associated with the estimated quantity, i.e., the chronological age. This distribution is then used to compute the probability that the examined individual is younger or older than an age threshold of interest, such as the age of majority [12], [13], [14]. The practical application of Bayesian methods can be facilitated to a great extent by using specific probabilistic graphical tools, such as Bayesian networks [15]. Bayesian networks combine elements of both graph and probability theories. The first one is applied to qualitatively define the structure of the model, taking into account the variables of interest (represented by nodes) and their respective probabilistic relationships (represented by directed arcs). The probability theory is then applied to define the nature of these relationships and to quantify their strength [15], [16]. Recently, a Bayesian network for age estimation was presented in scientific literature [14]. The network allows users to take into account evidence related to the degree of maturity observed in an examined individual and to provide both posterior probability distribution on the chronological age, as well as the posterior probabilities that the individual is older or younger than a given age threshold.

The aim of this paper is to study the performances of the Bayesian network in classifying individuals as younger or older than 18 years of age (i.e., the age of majority in numerous countries worldwide) on the basis of data related to the medial clavicular epiphysis development. Therefore, misclassifications (i.e., the rate of false minors and false majors) were investigated. Furthermore, the influence of prior probabilities assignment was also inspected. Analyses were performed using a sample available in scientific literature [17]. Note that the sample cannot serve as reference database and it was used only for exploratory analysis.

Section snippets

Material

For the analysis exposed in this paper, the data sample presented by Kreitner et al. was used [17]. The sample contains 380 European white subjects (males and females) under the age of 30 years lacking of development disorders. Each subject is described by his or her age and the classification of the degree of maturity reached by the medial clavicular epiphysis. The degree of maturity was assessed by an experienced radiologist according to a traditional four-stage classification [17], [18] by

Methods

Analyses were carried out with the Bayesian network for age estimation [14]. The structure of the network considers three variables: the chronological age (node Age in Fig. 1), the degree of maturity reached by the medial clavicular epiphysis, expressed in form of the assigned developmental stage (node Stage in Fig. 1) and the event that the examined individual is younger or older than 18 years of age. This latter is expressed in the form of two mutually exclusive propositions, P1: the examined

Performances in classification

Fig. 3 shows the distribution of the ORPost obtained with the model (e.g. Bayesian network). Descriptive statistics of these distributions are shown in Table 3.

The model well discriminates the two propositions of interest avoiding in that way the existence of ambiguous evaluative situations. This is shown in the histograms of Fig. 3, where all distributions of ORPost, given a particular stage, take a specific range of value with no overlapping. This means that the ORPost always corroborates one

Discussion

A sample related to the ossification status of the medial collar bone epiphysis was used for a preliminary investigation of the efficiency of the Bayesian network proposed in [14] for dealing with the classification of individuals (older or younger than 18 years of age). The development of the clavicle can be described by a few stages [18], [27], [28], making this physical attribute more suitable for a probabilistic classification of individuals according to a given age threshold rather than

Conclusion

In this paper, the performance of a probabilistic Bayesian method was tested in classifying young individuals as majors or minors based on their ossification status observed in the medial clavicular epiphysis. The Bayesian method was applied by means of the Bayesian network presented in [14]. Due to the nature of the data sample employed for the analyses, these latter are only explorative, however, the results obtained are promising, considering the low rate of misclassification encountered.

Acknowledgements

The authors wish to thank Ph.D. Cyril Muehlethaler for the proofreading. They are also grateful to the anonymous reviewers for their valuable comments.

References (31)

  • A. Schmeling et al.

    Practical imaging techniques for age evaluation

  • J. Taylor et al.

    Age evaluation and odontology in the living

  • S.L. Hackman et al.

    Age evaluation from the skeleton

  • D. Lucy

    The presentation of results and statistics for legal purposes

  • P.W. Thevissen et al.

    Human dental age estimation using third molar developmental stages: does a Bayesian approach outperform regression models to discriminate between juveniles and adults?

    Int. J. Legal Med.

    (2010)
  • Cited by (11)

    • An evaluation of statistical models for age estimation and the assessment of the 18-year threshold using conventional pelvic radiographs

      2020, Forensic Science International
      Citation Excerpt :

      Linear regression models were also used for age estimation [6,19], which were inappropriate for ordinal variables [20,21]. Some researchers have applied more sophisticated variable selection techniques, such as support vector machine (SVM) [22], decision tree [23], logistic regression [23], transition analysis [24] and Bayesian networks [25], to estimate age with better accuracy. However, few models have been tested on the pelvis.

    • Age estimation of living persons: A coherent approach to inference and decision

      2020, Statistics and Probability in Forensic Anthropology
    • Bayesian networks of age estimation and classification based on dental evidence: A study on the third molar mineralization

      2018, Journal of Forensic and Legal Medicine
      Citation Excerpt :

      This result can be achieved thanks to the intuitive nature of the CPTs linked to each node in the networks that allow users to entry and modify conditional and unconditional probabilities in an easy and intuitive way. For instance, the likelihood function for the evaluation of evidence form a single tooth was assessed in the present study by applying an unrestricted cumulative probit model to the reference sample, but other ordinal regression models have already been employed for age estimation,19,45 even in Bayesian networks13,43; ease the concrete computations in the inferential framework, since the probabilistic tool automatically provides the outcomes of interest avoiding the user to have to deal with the computational complexity of the algebraic handling of Bayes' theorem;

    • Ink dating part II: Interpretation of results in a legal perspective

      2018, Science and Justice
      Citation Excerpt :

      Until now, such a model has never been tested for ink dating using real reference data, only subjective probabilities were used to show the potential of this interpretation approach [2,19,30]. Different models were proposed in the literature, including Baysenets, for different ageing problematics such as evaluating the time since discharge [23], the moment of deposition of fingermarks [24], or the age of living people [25,26]. Each model has to take into account specificities such as the type of hypotheses (e.g. punctual times versus intervals), the type of data (e.g. continuous or discreet, uni or multivariate) and the type of ageing processes (e.g. regression fits and factors influencing transfer and influence storage).

    • Towards a Bayesian evaluation of features in questioned handwritten signatures

      2017, Science and Justice
      Citation Excerpt :

      As a consequence, information I is usually omitted from explicit graphical representation in Bayesian networks. Bayesian networks are very flexible, and have been used to support evaluative reasoning in very different forensic branches such as firearms [5], printed documents [6], signatures [7], forensic medicine [8] and DNA [9]. A review on the usage of Bayesian networks in forensic science can be found in Ref. [10].

    View all citing articles on Scopus
    View full text